DROPSONDE OBSERVATIONS FOR SURVEILLANCE NEAR THE REGION (DOTSTAR): AN OVERVIEW

Chun-Chieh Wu1, Po-Hsiung Lin1, Sim Aberson2, Tien-Chiang Yeh3, Wei-Peng Huang1, Jing-Shan Hong3, Guo-Chen Lu3, Kuan-Chien Hsu1, I-I Lin1, Kun-Hsuan Chou1, Pay-Liam Lin4, Ching-Hwang Liu5 Dept. of Atmospheric Sciences, National Taiwan University, , Taiwan 2Hurricane Research Division, NOAA/AOML, USA 3Central Weather Bureau, Taipei, Taiwan 4Department of Atmospheric Sciences, National Central University, Chung-Li, Taiwan 5Department of Atmospheric Sciences, Chinese Culture University, Taipei, Taiwan

ABSTRACT significant rainfall they bring is also a crucial water resource in Taiwan. Taiwan has been affected DOTSTAR (Dropwindsonde Observations for severely by many in recent years, and the Typhoon Surveillance near the Taiwan Region) is an loss of life and property has been staggering. The international research program conducted by typhoons that battered Taiwan in 2001 alone caused meteorologists in Taiwan partnered with scientists at 583 deaths, and more than US$ 400 million in the Hurricane Research Division (HRD) and the agricultural losses, nearly paralyzed the Taipei Rapid National Centers for Environmental Prediction (NCEP) Transit System, and did tremendous damage to of the National Oceanic and Atmospheric private and public sectors. Administration (NOAA). The experiment is based on successful surveillance missions conducted in the Prompted by their sense of social responsibility, Atlantic with NOAA's Gulfstream-IV jet aircraft. and the National Science Council’s (NSC) emphasis During the experiment, GPS dropwindsondes are on typhoon research, atmospheric science released from a jet aircraft flying above 42000 ft in and researchers in Taiwan initiated an interagency around tropical cyclones approaching Taiwan to research project on typhoons in September, 2001. collect critical meteorological data for improving the The “National Priority Typhoon Research Project” was analysis and the prediction of typhoons. formed in July, 2002, and the NSC approved the necessary 3-year funding. One key item of this After one-year of training, development and project involves a field experiment, Dropwindsonde installation of all the needed software and hardware in Observations for Typhoon Surveillance near the the aircraft, the DOTSTAR research team initiated Taiwan Region (DOTSTAR), which marks the typhoon surveillance in 2003. Two missions (in beginning of a new era for the surveillance of tropical Typhoons Dujuan and Melor) were conducted cyclones in the western North Pacific using GPS successfully,and seven or eight missions are expected dropwindsondes. to be conducted annually during the 2004 and 2005 typhoon seasons. 2. OVERVIEW of DOTSTAR The current manuscript provides an overview of a. Background the scientific objectives of DOTSTAR including Other than satellite observations, there has operational plans, organization, data management, been an unfortunate lack of observations in TCs in the and data archiving. Preliminary results of the two western North Pacific (NW Pacific), especially since missions in the first season in 2003 are presented. the U.S. discontinued typhoon reconnaissance flights The experiment marks the beginning of typhoon in the region in 1987. As described in Wu and Kuo surveillance in the western North Pacific and is (1999), the currently available data are not adequate expected to yield impressive improvements in typhoon enough to provide accurate initial and boundary research, observations and forecasting. conditions for the analyses used by numerical models 1. INTRODUCTION in TC forecasting. This deficiency puts a huge constraint on the accuracy of the TC forecasts, as well Typhoons are one of the most destructive as the general understanding of TCs in the NW Pacific weather systems in nature, and the most catastrophic region. weather phenomenon in Taiwan. Ironically the Considering the potential of dropwindsonde data in improving TC forecasting and understanding of * Corresponding author address: Dr. Chun-Chieh, Wu, their behavior, the DOTSTAR project was launched in Dept. of Atmospheric Sciences, National Taiwan 2002. The initiative of DOTSTAR is a collaborative University, No. 1, Sec. 4, Roosevelt Rd., Taipei, 106, effort between researchers from the National Taiwan Taiwan. e-mail: [email protected] University (NTU), Central Weather Bureau (CWB), in optimal (targeted) observation strategies for improving partnership with scientists at HRD and NCEP, building forecasts. upon work pioneered at NOAA's HRD to improve track 3) To validate/calibrate remote sensing data forecasts for TCs (Burpee et al. 1996; Aberson 2003). (such as the satellite- and radar-derived products) The key to DOTSTAR is the use of GPS around typhoons and help explore typhoon dynamics, dropwindsondes released from a jet aircraft flying such as the storm’s asymmetric structure, the above 42,000 feet in the environments of TCs that boundary-layer structure, and the typhoon-ocean approach Taiwan (Fig. 1). These sensors measure interaction. temperature, humidity, pressure, and wind velocity 4) To improve data assimilation. twice each second as they fall to the surface. c. Organization Information from the surveillance flights is transmitted by a satellite communication aboard the aircraft in real The Principal Investigator (PI) and director of time to the CWB (Fig. 2). To make maximum use of DOTSTAR is the first author of this paper, and the the data, the dropwindsonde data are assimilated in second to fourth authors serve as CO-PIs. The real time into the numerical models of CWB (i.e., the organization of DOTSTAR falls into two working global model, C-GFS), NCEP (i.e., the global model, groups. The observation team is in charge of the GFS; and the GFDL hurricane model), the U.S. Navy’s aircraft surveillance, including the creation of flight Fleet Numerical Meteorology and Oceanography tracks, the release of dropwindsondes, the data Center (FNMOC) (i.e., the global model, NOGAPS; communication, the processing of data by the the regional model, COAMPS; and the Navy version Airborne Vertical Atmosphere Profiling System of the GFDL hurricane model, GFDN) and the (AVAPS), the transmission of the data to the CWB Japanese Meteorological Agency (JMA, beginning in through the satellite phone, and then to NCEP, the 2004 season). The data are expected to enable FNMOC, and JMA, and the on-board data us to forecast storm track and intensity much better debugging/analysis and coding. The analysis and than without the additional data. research team is in charge of the receipt, analysis, and simulation/assimilation of the dropwindsonde data, The typhoon surveillance missions are carried assessment of the influence of the dropwindsonde out by an Astra SPX jet from the Aerospace Industrial data on typhoon forecasts, and all other follow-up Development Corporation (AIDC) in , Taiwan. research. The aircraft can cruise at abut 750 km h-1 and reaches a maximum height of 45,000 ft. It is the first During the 2002 hurricane season, four time during the past 16 years that aircraft have been researchers from Taiwan worked with HRD scientists used in the northwestern Pacific to routinely observe during operational surveillance missions, learning the typhoons. Each flight lasts for up to six hours during operational and scientific aspects of aircraft use to which time dropwindsondes are released about 150 to sample the hurricane environment. A number of 200 km apart, consistent with the spatial resolution of essential tasks were completed before the typhoon the traditional rawinsonde network. The flight route season in 2003. First, the whole aircraft platform, is designed to obtain observations targeted to the dropwindsonde equipment and the onboard data most sensitive region around the TC (i.e., the area receiving, analysis, and transmitting system and with the largest deep-layer-mean wind bred vectors programs were successfully set up at AIDC. Second, from the NCEP global ensemble forecasting system, the system for the realtime analysis and assimilation Aberson 2003), while modified to meet aircraft and of dropwindsonde data at CWB, as well as at NCEP air-control requirements. The project will enable and FNMOC was completed. Third, three test flights scientists to formulate future airborne observation and testing of the data flow and assimilation, were strategies, facilitate adaptive observations of typhoons, successfully performed from May to June, 2003. and improve data assimilation capabilities. The With all the above preparation work completed project is considered a pioneering step forward in in June, 2003, the DOTSTAR team was ready to basic research and forecasting of typhoons in the NW conduct surveillance missions. Pacific. 3. PRELIMINARY RESULTS b. Objectives Starting during the 2003 TC season, the ASTRA The objectives of the project are: jet was ready to fly missions in and around TCs in the 1) To conduct a pilot study to enhance NW Pacific near Taiwan. About 8 surveillance observations of the atmosphere and to improve the missions were expected for the first season. numerical guidance for NW Pacific TCs that may Nevertheless, because of the uncertainty of the TC affect the Taiwan area. track prediction and the constraints due to the two-day 2) To evaluate how the dropwindsonde data in-advance requirement for submitting flight plans to influence model track predictions, and study the the air traffic control agencies controlling different air spaces, only two missions were executed.1. Dujuan’s 250-km radius of gale-force wind, larger than On 1 September, 2003, the first DOTSTAR the value of 220 km estimated from satellite imagery mission was successfully completed around Typhoon at some operational centers with no knowledge of the Dujuan, with 11 dropwindsondes released (Fig. 3). dropwindsonde data. This information is particularly On 2 November, 2003, the second mission was important for issuing warnings for impacts from the conducted around , with 15 wind field as Dujuan approached land. dropwindsondes released (Fig. 4) and the ASTRA flew Although Dujuan did not make landfall on directly over the center of Melor. Note that a rather Taiwan, strong winds and heavy rain that it brought symmetric flight track was flown around the periphery caused extensive damage in southern Taiwan, of Dujuan, whereas in Melor, the aircraft passed over testifying to the large extent of the typhoon's wind field. the center of the typhoon and did not completely Dujuan continued to develop throughout 2 September, surround it with dropwindsonde data. The and ultimately caused widespread destruction in Hong descriptions of the two missions, along with the Kong, Macao, and Shantou in China's Guangdong preliminary results are described as below. Province. a. 3) Impact of the data on the numerical models 1) Synopsis Due to the lack of observations around TCs, TC After a two-month wait, the first DOTSTAR structure generally is not well represented in global mission was launched around Typhoon Dujuan analyses without the use of synthetic data. For between 0430 and 0800 UTC 1 September, Dujuan, the cyclonic winds in the middle and upper commencing a new milestone for typhoon troposphere are much stronger in the model initial observations and research in the NW Pacific. After conditions when the dropwindsonde are assimilated taking off from Taichung at 0430 UTC, the aircraft (Fig. 6). In particular, the closed circulation at high followed the northwestern edge of the peripheral levels associated with Dujuan is not analyzed by circulation of Typhoon Dujuan at an altitude of 41,000 CWB’s Global Forecasting System (C-GFS) (Fig. 6b) ft, and released its first dropwindsonde at 0520 UTC. without the dropwindsonde data. Nevertheless, after The aircraft released a total of 11 dropwindsondes at assimilating the dropwindsonde data, the wind intervals of about 200 km and at 200-250-km radius velocities are improved distinctly (Fig. 6a). The from the storm center (Fig. 3). The CWB began circulation center of C-GFS is located about 1 degree receiving data at 0630 UTC, and finished receiving all to the south of the actual location by the assimilation the dropwindsonde data before the plane landed in of the dropwindsonde data, likely due to both the poor Taichung at 0800 UTC. The CWB used the Internet first guess (the C-GFS did not perform the vortex to transmit data in real time to NCEP, and then relocation as in NCEP GFS) and insufficient model FNMOC, enabling all these centers to assimilate the resolution. dropwindsonde data into their 0600 UTC assimilation Figure 7a (7b; 7c) shows the differences in the cycles. initial deep-layer-mean (925-250 hPa) wind fields of 2) Analysis of the dropwindsonde data the C-GFS (NCEP GFS; FNMOC NOGAPS) with and without assimilating the dropwindsonde data. The The wind fields (Fig. 5) from the dropwindsonde region with the largest difference is collocated well data show that the cyclonic circulation was with the location of the dropwindsondes, with the asymmetric (partly because the aircraft did not fly maximum value of about 7 m s-1 for GFS, 3 m s-1 for around the storm at a constant radius from the center) GFS and 7 m s-1 for NOGAPS. Due to some and extended from 925 hPa (with a maximum wind of technical problems, only three of the eleven about 40 m s-1 in Fig. 5a) to 200 hPa (with a maximum dropwindsondes were assimilated into NCEP GFS in of 25 m s-1 in Fig. 5c). The data revealed that realtime.2 The difference of the deep-layer-mean Typhoon Dujuan was stronger and more coherent wind in GFS is in good agreement with Aberson vertically than had been analyzed without the (2003), though it is not clear why the the C-GFS dropwindsonde data at CWB (not shown), as well as (NOGAPS) has much larger differences and larger at the Joint Typhoon Warning Center (JTWC). This impact regions to the south (southeast) of the data is the first time ever that this type of valuable locations. information was available to the forecasters at CWB in realtime. The data also led to a better estimation of The comparison of the model runs of Dujuan

1 This drawback will be improved upon in subsequent 2 This problem has been fixed though a rerun seasons, as a better and more flexible flight operation incorporating all 11 dropwindsondes has not been application procedure to the air-control agencies in the completed, mainly due to the heavy operational load and has been set up. at NCEP. with and without the dropwindsonde data show that depicted along the leg (drops 3 to 7, Fig. 4) when the the NCEP GFS forecasts are improved by 32% to aircraft flew directly over the center. Note that the 81% between 6 and 30 h (Table 1 and Fig. 8) and by apparent warm core exists in the center (Fig. 9), with less than 10% beyond 36h. The average the saturated equivalent potential temperature higher improvement from 6 to 48 h is 35% (note that only than that of the surroundings by about 5-10 degrees. three dropwindsondes were assimilated). Some Figure 9 shows that Melor has a rather large with improvement of about 25% (not shown) between the radius of about 100 km, about the size estimated 6-24-h can also be identified in the NCEP GFS from the satellite imagery (Fig. 4). The axis of the initialized at 1200 UTC 1 September (i.e., the maximum wind speed tilts outward with height, with a dropwindsonde information were carried into the next maximum measured wind speed of 24 m s-1 at assimilation cycle through the 6-h forecasts from 0600 900-hPa near the 6th dropwindsonde. Note that this UTC 1 September). is also the first time that this type of detailed inner However, no impact on the NCEP GFDL structure of TCs in the NW Pacific is available in real hurricane model track forecast is found (not shown). time at CWB. The GFDL model does not assimilate the 3) The impact of the dropwindsonde data to dropwindsonde data directly, but uses the initial fields numerical models from the NCEP GFS. It is possible that the GFDL The track forecasts for Typhoon Melor shows synthetic vortex removed the dropwindsonde one of the greatest challenges for numerical models in information from the NCEP GFS in this case. On the the typhoon season of 2003, as most models other hand, the impact of the dropwindsonde data predicted that Melor would head into the South China from all eleven dropwindsondes on NOGAPS forecast Sea after passing through Luzon. However, Melor (Table 1) is completely different from that of the NCEP turned northward toward Taiwan (Fig. 10). For this GFS. The impact on track forecasts is negative case, the NCEP GFS model forecasts show negative during the first 24h, then positive at 36 and 48h. impact with the dropwindsonde data (Fig. 10 and Follow-up analyses on the physical features leading to Table 2). Significant degradations (about 50% in the such improvement in the NCEP GFS and degradation first 24 h, Table 2) are seen from 6 to 36 h. This in NOGAPS are ongoing through use of potential result is in agreement with Aberson (2003) who vorticity diagnosis (Wu et al. 2003, 2004). It is hoped showed that the sampling of the entire target feature is that such analyses may provide insight into how the needed to improve the forecasts, otherwise inclusion of the dropwindsonde data affects the degradation is probable. performance of each model. The initial results of DOTSTAR indicate a b. Typhoon Melor golden opportunity for improving the track prediction 1) Synopsis of typhoons in the western North-Pacific (near Taiwan). The observations for Typhoon Melor were More dropwindsondes will be released into the obtained between 0400 and 0700 UTC 2 November, periphery of typhoons near Taiwan during the typhoon as Melor crossed Luzon and headed northward seasons of 2004 and 2005. As the number of toward the southern tip of Taiwan. During this observations increases, we expect to see a more mission, the jet departed from Taichung at 0400 UTC, statistically significant evaluation of the impact of rounded northern Taiwan, and flew south along the these dropwindsondes on TC track predictions. east coast of Taiwan. The aircraft released the first While it will take careful assessment of the results of GPS dropwindsonde near Green Island, passed the this project to show whether the dropwindsonde data southern tip of Taiwan, and dropped several more can be used to improve typhoon forecasting, it is likely dropwindsondes as it continued on a southerly course. that these valuable data will lead to major Flying at 41,000 feet, the aircraft successfully overflew breakthroughs in typhoon research. the center of Typhoon Melor and gathered important Meanwhile, by flying over a typhoon eye for the data on the structure of the typhoon near its eye (Fig. first time, the mission for Melor also laid the 9). The aircraft immediately turned toward the west groundwork for future observations of TC’s inner core upon arriving at the northern tip of Luzon. After structure. The pilots from AIDC are scheduled to visit rounding the western edge of the typhoon (this flight NOAA/AOC during the spring of 2004 to gain pattern was required due to the air control issues) and experience in flying in the storm (hurricane) core. releasing more dropwindsondes, the aircraft flew back These exchanges should be quite positive for future to the southwestern tip of Taiwan, released its 15th DOTSTAR missions. Meanwhile, work is ongoing to and final dropwindsonde, and returned to Taichung at compare the data collected by the mission with 0700 UTC. satellite data and data from the Doppler radar station The analysis of the dropwindsonde data at Kenting (Lee et al. 2000), which is located near the southern tip of Taiwan. This may shed new light on The detailed structure of Melor can be the study of TC structure, rainbands, and circulation. NOGAPS singular vector analysis (Melinda Peng, personal communication 2004). 4. SUMMARY AND FUTURE PLANS . The research group maintains close In light of the heavy damage done by typhoons communication and exchanges important scientific to Taiwan each year, the NSC of Taiwan places a ideas with HRD/NOAA, while gaining helpful technical great premium on typhoon research, and therefore experience. As the DOTSTAR research team has appropriated US$ 1 million for the "National continues to harvest important data and gain valuable Priority Typhoon Research Project” each year for experience, we believe that future typhoon three years (from August 1, 2002 to July 31, 2005), observations will reach full maturity, enabling especially for the field experiment, “Dropwindsonde significant progress in both academic research and Observations for Typhoon Surveillance near the typhoon forecasting. For example, because the NW Taiwan Region (DOTSTAR)”. DOTSTAR is an Pacific is the region with the most frequent and international research program conducted by intense TCs worldwide, with the very high meteorologists in Taiwan, partnered with scientists at vertical-resolution observations from the NOAA HRD and NCEP. This project marks the dropwindsondes, DOTSTAR offers a very good beginning of a new era for aircraft surveillance of opportunity for the detailed measurement of the typhoons in the western North Pacific. boundary layer wind and air-sea exchange coefficient Built upon work pioneered at NOAA's HRD, the at high wind conditions (Powell et al. 2003), which are key to the project is the use of airborne sensors, also one critical element for improving our dropwindsondes, which are released from jet aircraft understanding of the TC intensity change (Emanuel flying above 42,000 feet in the environment of a 1999; Wang and Wu 2003). Many of the wind and . These sensors gather temperature, moisture data in the entire troposphere can also prove humidity, pressure, and wind velocity information to be a unique dataset for the validation and every half second as they fall to the surface. calibration of many remotely sensed data for TCs in Information from the surveillance flights is transmitted the NW Pacific region. in near realtime to the CWB of Taiwan, as well as to It is hoped that DOTSTAR will shed light on NCEP and FNMOC. Starting in the 2004 season, the typhoon dynamics, enhance typhoon track forecasting data are also scheduled to be transmitted to JMA in accuracy, place Taiwan at the forefront of international real time. The data are assimilated operationally into typhoon research, and make a significant contribution the numerical models of CWB, NCEP (GFS/GFDL) to the study of typhoons in the northwestern Pacific and FNMOC (NOGAPS/COAMPS/GFDN). and East Asia region. DOTSTAR is expected to provide valuable data that can help increase the accuracy of TC analyses and ACKNOWLEDGMENTS track forecasts, to assess the impact of the data on The authors gratefully acknowledge Dr. numerical models, to evaluate the strategies for Hung-Duen Yang, Director of the Division of Natural adaptive observations, to validate/calibrate remotely Sciences of National Science Council of Taiwan, for sensed data, and to improve our understanding of TC his strong support of the DOTSTAR program. dynamics. Special thanks go to Dr. Ching-Yen Tsay, Dr. Shaw Liu, On September 1, 2003, the first DOTSTAR Mr. Hsinn-Liang Shieh (and CWB), and Dr. Frank mission was successfully completed around Typhoon Marks (and Hurricane Research Division) for their Dujuan. On November 2, the second mission was help and advice. Helpful discussions with Drs. Hugh launched, and the aircraft flew over the center of Willoughby, Kerry Emanuel, Yoshio Kurihara, Russ Typhoon Melor. Preliminary results have shown that Elsberry, T. N. Krishnamurti, Greg Holland, Kuo-Nan these observations have provided helpful data for the Liou, Michael Montgomery, Simon Chang, C-P Chang, analysis, prediction and understanding of both Dujuan Bill Kuo, Wen-Chau Lee, Yuqing Wang, Lou Lee, and Melor. George Chen, Gen-Rong Liu, Ben Jou, Chung-Hsin We are expecting to undertake at least 8 Sui, and Hung-Chi Kuo are appreciated. We surveillance missions in both 2004 and 2005. particularly thank the nice contribution from all the Instead of using the previous targeting strategy, where team members of the DOTSTAR project and the the dropwindsondes are released at locations in which collaborators at AIDC, especially from Chia-Hsiang the “spread” (or standard deviation) of an ensemble Tsao, Li-Shyong Yin and Ying-Hao Chen, who were on forecast from NCEP GFS is large at the observation board of both surveillance flight missions in Dujuan time, we plan to examine the new targeting strategy and Melor. Important help from the Japanese Civil based on the Ensemble Transform Kalman Filter Aviation Bureau and the Meteorological Research (Majumdar et al. (2002), which predicts the signal Institute (MRI)/JMA (Dr. Tetsuo Nakazawa), Air variance (reduction in forecast error variance) for all Transportation Office of Philippines, NCEP (Drs. feasible deployments of targeted observation, or the Hua-Lu Pan and Naomi Surgi), GFDL (Morris Bender) and FNMOC (Drs. Mary Alice Rennick and Brain of Typhoon Alex (1987). Mon. Wea. Rev., 128, Strahl) are also appreciated. Another special thanks 3982-4001. to Dennis Keyser of NCEP/EMC for enabling the GFS Majumdar, S. J., C. H. Bishop, B. J. Etherton and Z. rerun of the Melor mission and for help in debugging Toth, 2002: Adaptive sampling with the communications. Thanks also goes to Russ Treadon Ensemble Transform Kalman Filter. Part II: and Mark Iredell of NCEP/EMC and Timothy Marchok Field program implementation. Mon. Wea. of GFDL for help in running the NCEP models, and to Rev., 130, 1356-1369. Dr. Stephen Lord and the staff of the NCEP Central Powell, M. D., P. J. Vickery, and T. A. Reinhold, 2003: Computing Facility who enabled the runs to proceed Reduced drag coefficient for high wind speeds in real time. The research is supported by Grant in tropical cyclones. Nature, 422, 279-283. NSC 91-2119-M-002-032 and NSC Wang Y., and C.-C. Wu, 2003: Current understanding 92-2119-M-002-009-AP1. of tropical cyclone structure and intensity changes - A review. Meteor. and Atmos. REFERENCES Physics (in press). Aberson, S. D., 2003: Targeted observations to Wu, C.-C., and Y.-H. Kuo, 1999: Typhoons affecting improve operational tropical cyclone track Taiwan: Current understanding and future forecast guidance. Mon. Wea. Rev., 131, challenges. Bull. Amer. Met. Soc., 80, 67-80. 1613-1628. Wu, C.-C., T.-S. Huang, W.-P. Huang, and K.-H. Chou, Burpee, R.W., J. L. Franklin, S. J. Lord, R. E. Tuleya, 2003: A new look at the binary interaction: and S. D. Aberson, 1996: The impact of Omega Potential vorticity diagnosis of the unusual dropwindsondes on operational hurricane and southward movement of Typhoon Bopha (2000) track forecast models. Bull. Amer. Met. Soc., 77, and its interaction with Typhoon Saomai (2000). 925-933. Mon. Wea. Rev., 131, 1289-1300. Emanuel, K. A., 1999: Thermodynamic control of Wu, C.-C., T.-S. Huang, and K.-H. Chou, 2004: hurricane intensity. Nature, 401, 665-669. Potential vorticity diagnosis of the key factors Lee, W.-C., B. J.-D. Jou, P.-L. Chang, and F. D. Marks, affecting the motion of Typhoon Sinlaku (2002), 2000: Tropical cyclone kinematic structure Mon. Wea. Rev. (in press). retrieved from single-Doppler radar observations. Part III: Evolution and structure

Table 1. Forecast track errors (in km) from the NCEP GFS and FNMOC NOGAPS models (initialized at 0600 UTC 1 September) with and without the dropwindsonde data, and the relative improvement (%).

Table 2. Forecast track errors (in km) from the NCEP GFS model (initialized at 0600 UTC 2 November) with and without the dropwindsonde data, and the relative improvement (%).

Fig. 1. The area (shaded) for proposed typhoon surveillance in DOTSTAR.

Fig. 2. Data flow chart for the aircraft data.

Fig. 3. GMS-5 Visible Imagery at 0525 UTC 1 Fig. 4. The GMS-5 Visible Imagery at 0625 UTC 2 September 2003, and the dropwindsonde locations November 2003, and the dropwindsonde locations (solid dots) for Typhoon Dujuan, 0430 – 0800 UTC 1 (solid dots) for Typhoon Melor, 0400 – 0700 UTC 2 September 2003. The numbers indicate the November, 2003. The numbers indicate the sequence of the dropwindsondes released. sequence of the dropwindsondes released.

Fig. 5. The wind fields (one full wind barb represents 10 m s-1) from the dropwindsonde data for Typhoon Dujuan at 0600 UTC 1 September 2003, at (a) 925 hPa, (b) 500 hPa, and (c) 200 hPa.

Fig. 6. The initial 300-hPa wind fields (one full wind barb represents 10 m s-1) of CWB’s GFS (a) with assimilation of the dropwindsonde data, and (b) without assimilation of the dropwindsonde data. The 300-hPa winds from the dropwindsonde observations are shown in bold.

Fig. 7. The difference of the initial deep-layer-mean (925-250-hPa) wind in the model with and without assimilation of the dropwindsonde data in (a) NCEP GFS, and (b) FNMOC NOGAPS models. The contour interval is 1 m s-1. The X’s show the locations of the dropwindsonde data being assimilated.

Fig. 8. The best track (BTRACKS) and the corresponding 48-h model forecast tracks of Typhoon Dujuan from the NCEP GFS (initialized at 0600 UTC 1 September) with the assimilation of the dropwindsonde data [GFS(3drop), in circles for every 6 h] and without assimilation of the dropwindsonde data [GFS(nodrop), in dots for every 6 h].

Fig. 9. The vertical cross section of wind speed (with a contour interval of 3 m s-1; one full wind barb represents 10 m s -1) and the saturated equivalent potential temperature (shaded) from the third to the seventh drops (from the north to the south, as shown in Fig. 5) of Typhoon Melor.

Fig. 10. The best track (BTRACKS) and the corresponding 48-h model forecast tracks of Typhoon Melor from the NCEP GFS (initialized at 0600 UTC 2 November) with assimilation of the dropwindsonde [GFS(alldrop), in circles for every 6 h] data and without assimilation of the dropwindsonde data [GFS(nodrop), in dots for every 6 h].